Study And Implementation Of Positioning And Moving Pattern Recognition Of Indoor Pedestrian Based On Mobile Phone Sensors

This thesis mainly covers a positioning algorithm,based on mobile phone sensors,and a moving pattern recognition approach of indoor pedestrian.About the indoor positioning,it presents a set of algorithms for synchronous trajectory mapping and positioning of a pedestrian,which gets and manipulates the data acquired through inertial sensors and a WiFi detector embedded in a mobile phone.For the pedestrian trajectory mapping,it analyses the data from tri-axial accelerometer to identify pedestrian's steps,and records the starting and ending time of each step.Using quaternion method,it calculates the data from the tri-axial gyroscope to determine the gesture of the phone and the moving direction of the pedestrian who holds the phone.It employs a novel WiFi adaptor location(landmark)marking strategy,which identifies marks through the analysis of signal strength change,to eliminate the marking errors caused by traditional marking model,which is vulnerable to the environmental complexity and the inconformity of WiFi detectors.Implementing a particle filter in the trajectory mapping is effective to upgrade the mapping precision.It uses the updates of pedestrian's gait and moving direction,determined by checking the data from inertial sensors,to drive the algorithm's iteration,and,along the moving track,detects and marks WiFi landmarks in the grid map automatically.Taking the distance between particle and WiFi landmark as the posterior information of particle filtering,WiFi landmarks and pedestrian trajectory adjust themselves mutually.The synchronous mapping and positioning goes through without knowing indoor structure of the building beforehand.Experimental results show a better fitting between the mathematically mapped trajectory and the physical trajectory.The positioning error is less than 5 meters.The fast positioning algorithm mainly serves identifying the different floors of a multi-storey building and predicting pedestrian's coordinates on a specific floor automatically.It collects and reconstructs the data generated during the simultaneous mapping and positioning,then learns and trains models through machine-learning process.Sorting out the acquired WiFi data corresponding to each floor,applying logistic regression to the model training,it obtains a floor recognition model,which works well to identify the floor where the phone is,depending on current WiFi signals.It builds an equal-sized grid for each floor,which has already been pedestrian-trajectory-mapped.The grid has columns and rows of cells,each cell has a unique coordinates.Matching each cell's position and its corresponding WiFi signals,harnessing supportive vector machines,it optimizes a model to solve out the cell's coordinates with the real-time WiFi data as input.About the recognition of moving pattern,it provides an algorithm based on the measurements from mobile phone's inertial sensors and machine-learning classifiers.By acquiring and marking the inertial data in a variety of motion types,and extracting features according to the feature definition rules,and applying machine-learning method to train the classification model,it demonstrates a practical path to identify the moving pattern of resting,walking or running.